This sort of conversation is usually pretty subjective so it will be impossible to actually come to any agreement.
That said, ignoring the JVM in this conversation is near criminal. High performance GC, JIT, standardized profiling, etc. are all major steps forward. Are they language improvements? No, but they proved that a VM is a viable target platform and that was under serious debate in the late 90s.
I ignored the JVM because it really isn't that special at all, and its supposed benefits haven't been proven to exist in reality.
Various types of VMs predated it by decades, including ones that used various forms of GC and JIT compilation. See some of the Pascal and Smalltalk implementations from decades ago as an example of these systems. These did see a fair amount of use, in practice.
The JVM isn't particularly portable. It may support most of the major mainstream OSes used today, but it's not like portable C or C++ code (including other language implementations for Perl, Python, and so forth) that can run on all sorts of obscure and ancient systems.
Its performance isn't particularly remarkable, either, even with all of the effort that some very large companies have put into it. Its been pretty much relegated to server-side use at this point, and even then we're seeing more and more effort made to move away from it. People are realizing that it's often better to go native, as we're seeing with newer languages like Go and Rust.
The JVM did get a lot of hype, and has seen a lot of use, but it's mediocre at best. There's nothing particularly earth-shattering about it.
That said, ignoring the JVM in this conversation is near criminal. High performance GC, JIT, standardized profiling, etc. are all major steps forward. Are they language improvements? No, but they proved that a VM is a viable target platform and that was under serious debate in the late 90s.